Action recognition is a widely studied problem and many solutions have been introduced over the years. However, classification with weakly-labeled large scale web data continues to be a challenge due to the noisy content. In this study, we utilize a representation method that is based on selected distinctive exemplars, motivated by the success of methods that discover the discriminative parts of data for better classification. These exemplars are chosen to be representative from each category separately and called as "prototypes". After the selection, we use these prototypes to describe the entire dataset. Following the traditional supervised classification methods and utilizing the available state-of-the-art low and deep-level features, we show that even with simple selection and representation methods, use of prototypes can increase the recognition performance.